Short-term load forecasting based on CEEMDAN-FE-ISSA-LightGBM model
Zhihao Li, Zhongli Chen
Abstract
To address the problems of low load forecasting accuracy due to the strong non-stationarity of electric loads, this paper proposes a short-term load forecasting method based on a combination of the complete ensemble empirical modal decomposition adaptive noise method-fuzzy entropy (CEEMDAN-FE) and the Light Gradient Boosting Machine (LightGBM) optimized by the improved sparrow search algorithm (ISSA). First, the original data are decomposed by the complete ensemble empirical modal decomposition adaptive noise algorithm to obtain the eigenmodal components (IMFs) and residual values. Second, the obtained sequences are entropy reorganized by fuzzy entropy, and thus new sequences are obtained. Third, the new sequences are input into the improved sparrow search algorithm-Light Gradient Boosting Machine model for training and prediction. The improved sparrow search algorithm algorithm can realize parameter optimization of the Light Gradient Boosting Machine model to make the data match the model better, and the predicted values of each grouping of the model output are superimposed to obtain the final predicted values. Finally, the effect is compared by the error function, and the comparison results are used to test the performance of the algorithm. The experiments showed that the smallest evaluation metrics were obtained in Case 1 (MAE = 32.251, MAPE = 0.0114,RMSE = 42.386, R2 = 0.997) and Case2 (MAE = 3.866, MAPE = 0.003, RMSE = 5.940, R2 = 0.997).